When first language is not first: an functional magnetic resonance … › articles › prior › When first... · 2018-12-03 · COGNITIVE NEUROSCIENCE When ﬁrst language is not

COGNITIVE NEUROSCIENCE

When first language is not first: an functional magneticresonance imaging investigation of the neural basis ofdiglossia in Arabic

Michael Nevat,1,2 Asaid Khateb1,2,3 and Anat Prior2,31The Unit for the Study of Arabic Language, Edmond J. Safra Brain Research Center for the Study of Learning Disabilities,Faculty of Education, University of Haifa, Mount Carmel, Haifa 31905, Israel2The Laboratory for the Study of Bilingualism, Edmond J. Safra Brain Research Center for the Study of Learning Disabilities,Faculty of Education, University of Haifa, Haifa, Israel3Department of Learning Disabilities, Faculty of Education, University of Haifa, Haifa, Israel

Keywords: Arabic, bilingualism, healthy subjects, Hebrew, semantic categorization

Abstract

In Arabic, the language used for everyday conversation (‘spoken Arabic’ – SA) differs markedly from literary Arabic (LA), which isused for written communication and formal functions. This fact raises questions regarding the cognitive status of the two varietiesand their processing in the brain. Previous studies using auditory stimuli suggested that LA is processed by Arabic native speak-ers as a second language. The current study examined this issue in the visual modality. Functional magnetic resonance imaging(fMRI) responses were collected while Arabic–Hebrew bilinguals performed a semantic categorization task on visually presentedwords in LA, SA and Hebrew. Performance on LA was better than SA and Hebrew, which did not differ from each other. Activa-tion in SA was stronger than in LA in left inferior frontal, precentral, parietal and occipito-temporal regions, and stronger than inHebrew in left precentral and parietal regions. Activation in SA was also less lateralized than activation for LA and Hebrew, whichdid not differ from each other in terms of lateralization, though activation for Hebrew was more extensive in both hemispheresthan activation for LA. Altogether, these results indicate an advantage for LA in the current study, presumably due to participants’proficiency in reading in this language. Stronger activation for SA appears to be due to the relative unfamiliarity of written wordforms in SA, which could also explain differences in performance between the two languages. However, the stronger activationobserved in the left parietal cortex may also reflect stronger associations among words in SA.

Introduction

The Arabic language is a typical example of ‘diglossia’ (Ferguson,1959), which is a socio-linguistic situation in which the languageused for everyday conversation (i.e. spoken Arabic – SA), differsmarkedly from the written language (i.e. literary Arabic – LA, alsocalled modern standard Arabic). SA and LA varieties differ in termsof the age and manner of acquisition as well as their use. SA, thespoken local dialect, is the first language (L1) acquired by nativespeakers of Arabic, serves strictly for oral communication and doesnot typically exist in written form. LA, a highly codified form, isacquired later in childhood primarily through formal education,though children are differentially exposed to it aurally throughaudio-visual media (Abu-Rabia, 2000; Saiegh-Haddad, 2003; Bou-delaa & Marslen-Wilson, 2013). LA is used for reading and writingand formal speech functions (religious sermons, official speeches,news broadcasts and teaching; Ibrahim & Aharon-Peretz, 2005;

Levin et al., 2008; Saiegh-Haddad et al., 2011). LA differs fromSA in the phonological, morpho-syntactic and lexical-semanticdomains (Saiegh-Haddad et al., 2011). Also, whereas LA is homo-geneous across the Arabic-speaking world, considerable differencesexist between dialects of SA used in different regions (Ayari, 1996;Saiegh-Haddad, 2003, 2005; Saiegh-Haddad et al., 2011; Boudelaa& Marslen-Wilson, 2013).Implications of diglossia for the acquisition of basic reading

processes by Arabic speakers have repeatedly been discussed(Saiegh-Haddad et al., 2011), and the extent to which it resemblesbilingualism has been examined only behaviorally. For instance,Eviatar & Ibrahim (2000) reported that the performance of Arabic-speaking children on tests assessing meta-linguistic abilities resem-bled that of bilinguals. Also, Ibrahim & Aharon-Peretz (2005) inves-tigated the cognitive status of SA and LA (and Hebrew as a secondlanguage, L2) using semantic priming effects during auditory lexicaldecision. They reported that priming effects were larger when primeswere in SA and target words were either in LA or in Hebrew thanthe reverse (LA or Hebrew primes and SA targets). Furthermore, thepriming effects for LA and Hebrew were identical (Ibrahim, 2009).These effects resemble findings in bilinguals (Keatley et al., 1994;

Correspondence: Prof A. Khateb, 1The Unit for the Study of Arabic Language, asabove.E-mail: [email protected]

Received 7 April 2014, revised 9 June 2014, accepted 17 June 2014

© 2014 Federation of European Neuroscience Societies and John Wiley & Sons Ltd

European Journal of Neuroscience, Vol. 40, pp. 3387–3395, 2014 doi:10.1111/ejn.12673

Gollan et al., 1997), which show larger forward priming (from L1to L2) than backward priming (from L2 to L1), and supported theview that SA is cognitively represented as L1 and LA as L2 (Ibra-him & Aharon-Peretz, 2005).The current study assessed the neural basis of diglossia by analy-

sing the processing of visually presented LA and SA words in adultArabic speakers, and comparing both to the participants’ formal L2(Hebrew). Participants performed a semantic categorization task,previously shown to reliably activate left hemisphere language areas(Seghier et al., 2004, 2008). Functional magnetic resonance imaging(fMRI) and behavioral measures were analysed to investigatewhether separate neural processes might support visual word pro-cessing in the languages examined.Because the diglossic situation is unique in that it distinguishes

between an L1 for oral communication and a different L1 for liter-acy, stronger activation could be expected in areas differentiatingthe weaker from the dominant language (see examples in Warten-burger et al., 2003; Meschyan & Hernandez, 2006; R€uschemeyeret al., 2006) when comparing visual word processing in SA orHebrew to LA, with a performance advantage for LA.

Materials and methods

Participants

Twenty-six healthy women participated in the study. Due to atechnical malfunction, the scanning of one participant could notbe completed and she was therefore excluded from the analyses.The remaining participants’ age was between 18 years and 4 monthsand 23 years and 10 months (mean = 20.3 years, SD = 1.4 years).All participants were Arabic–Hebrew bilinguals, in their first year ofundergraduate studies in the Faculty of Education at the Universityof Haifa. Participation in the experiment counted as partial fulfill-ment of course credit. All participants were right-handed except forone [handedness laterality index (LI) mean = 58.6, SD = 18.7], hadnormal or corrected-to-normal vision, had no history of neurologicalor psychiatric disorders, and reported no use of any psychoactivemedication at the time of the experiment. Participants were informedabout the purposes of the study and gave written informed consentbefore participating. The study protocol was approved by the ethicscommittee of the University of Haifa and conforms with the WorldMedical Association Declaration of Helsinki.Concerning the participants’ language background, they had all

been exposed to formal instruction of LA since the first grade, andof Hebrew as an L2 since the third grade (at the age of 8–9 years).They had all completed their secondary school studies and success-fully passed the high-school matriculation exams with Hebrew asL2. In addition, all students were required to pass the Hebrew Profi-ciency Test (‘YAEL’ – National Institute for Testing & Evaluation)with a score of at least 115 (out of 150 possible points) as a prere-quisite for admission to the University, where they followed theircourses in Hebrew. Therefore, at the time of the experiment, allcould be considered as proficient bilinguals.

Stimuli and procedure

Participants performed a semantic categorization task based onSeghier et al. (2004), in which they were asked to judge whetherpairs of words presented on the screen were semantically related(SR; i.e. belonged to the same semantic category) or not. The stim-uli were high-frequency, concrete imageable nouns, in Arabic andHebrew. For the selection of the stimuli, a questionnaire was first

presented to a group of 30 native Arabic-speaking participants, whowere asked to rate the frequency of 220 SA words and 220 LAwords using a scale from 0 to 6 (0 unknown or least frequent, 6most frequent). All stimuli were selected so as to minimize phono-logical overlap between SA and LA words (but also with Hebrewtranslation equivalent). The most frequent 200 words in each lan-guage variety (LA – mean = 4.34, SD = 0.95; SA – mean = 4.75,SD = 1.17) were then retained for this study.From the selected words (length three-six letters), 50 SR word

pairs and 50 semantically unrelated (SU) word pairs were formedin each language (SA and LA). These pairs were then presented ina questionnaire to another group of Arabic-speaking participants,who rated semantic relatedness of the words in each pair using ascale from 0 to 5 (0 for least related and 5 for most related) toensure their suitability for the different conditions. The averagerelatedness in each list was above 3 for related pairs(LA = 3.87 � 0.38; SA = 3.60 � 0.18), and below 0.5 for unre-lated pairs (LA = 0.28 � 0.18; SA = 0.28 � 0.17). Finally, thereadability of these pairs was assessed in a pilot study conductedwith 10 adult participants using a computerized speeded semanticjudgment task (as in Khateb et al., 2003). An item-by-item analy-sis performed on the results of this study allowed selecting onlypairs that yielded at least 8/10 correct responses. For the Hebrewwords, the stimuli consisted of translation equivalents of wordsfrom SA and LA.The words in each pair were simultaneously displayed, one word

beneath the other, at the center of a computer screen located outsidethe scanner. Mirrors fastened to a head coil reflected the stimuli, sothat they could be viewed by the participants. We used a block para-digm that alternated between the semantic categorization and thecontrol condition. In the control condition, pairs of Greek characterstrings were simultaneously presented, and participants judgedwhether the strings were physically identical or not. In this condi-tion, which was used for all language blocs and mainly involvesvisual processing, the stimuli were designed so as to resemble thosein the semantic categorization blocks in terms of the number ofcharacters and the spatial extent.The same semantic categorization task was used for the three lan-

guages: one using stimuli in LA; one using stimuli in SA written inthe Arabic orthography; and one using stimuli in Hebrew. Becausethe participants were all skilled readers in Arabic and Hebrew, thewords in all language blocs were presented without diacritics(i.e. short vowels) as is customary for adult readers (Abu-Rabia,2001). The order of presentation of the language runs was counterbal-anced across participants. In each of the runs the activation conditionconsisted of 72 word pairs (48 SR pairs and 24 SU pairs) dividedinto six blocks. The subjects performed a yes/no task, and gave aresponse (using their left thumb) to indicate whether the two wordsin each pair were related or unrelated. The control condition also con-sisted of 48 pairs of identical Greek letter-strings and 24 pairs ofvisually different strings. Here, participants indicated whether thetwo strings in each pair were visually identical, or not. In all condi-tions (both activation and control) and language runs, stimulus pairswere presented every 2 s on the screen for 600 ms and in blocks of24 s, repeated six times per condition. Hence, alternating blocks ofactivation–control conditions yielded a total duration of 4.8 min perlanguage run. Responses were given using an MR-compatibleresponse box that allowed registering the performance of the subjectand the reaction times (RTs). In order to ensure full comprehensionof the task demands, participants were provided with instructionsbefore entering the scanner and underwent a training session of afew trials.

© 2014 Federation of European Neuroscience Societies and John Wiley & Sons LtdEuropean Journal of Neuroscience, 40, 3387–3395

3388 M. Nevat et al.

fMRI acquisition

The experiments were conducted using a 3T MRI scanner (GE Dis-covery MR750) at the Rambam medical center in Haifa. A high-res-olution T1-weighted anatomical scan was recorded for eachparticipant [voxel size – 1 9 1 9 1 mm; number of slices – 148;repetition time (TR) = 12.73 ms; echo time (TE) = 5.42 ms]. Foreach experimental run (task), 145 dynamic volumes with axial con-tiguous ascending acquisitions were recorded (voxel size –3.44 9 3.44 9 3.4; matrix size – 64 9 64; number of slices – 43;interslice gap – 0%; TR = 2000 ms; TE = 30 ms; field of view –220; flip angle = 60°). For each run, the functional scanning wasalways preceded by 10 s of dummy scans to insure tissue steady-state magnetization.

Whole-brain analysis

MRI data were analysed with the Statistical Parametric MappingSPM8 software (http://www.fil.ion.ucl.ac.uk/spm/). All functionalvolumes were subjected to standard preprocessing procedures(Friston et al., 2007), including: spatial realignment; normalization[to Montreal Neurological Institute (MNI) space with 2 9 2 9 2 mm3

voxel size]; and smoothing with an isotropic 5-mm full-width athalf-maximum Gaussian kernel. Time-series from each voxel werehigh-pass filtered (1/128 Hz cutoff). After preprocessing, analyses atthe level of the individual participant (first-level analysis) were per-formed using the general linear model applied to each voxel (Fristonet al., 1995; Worsley & Friston, 1995) and an auto-regressive [AR(1)] function to account for temporal correlations between themacross the whole brain. Each run was modeled as a distinct session,and each condition within a run (semantic categorization or control)was separately modeled. At the group level, a two-way (lan-guage 9 condition) ‘flexible factorial’ model was then specified,resulting in a total of six regressors (three languages 9 two condi-tions).Activation in all semantic categorization blocks was compared

with activation in all control blocks in order to identify all regionsthat were active during processing of stimuli in at least one of thelanguages. Afterwards, activation during semantic categorization ineach of the languages was compared with activation during controlcategorization in the same run. The resulting differences were thenentered into a conjunction analysis in order to identify regions activein all languages, and were then compared with each other in orderto examine differences in activation between languages. In these lat-ter comparisons, the results were masked with the activation mapobtained for each language minus its baseline, to ensure that theresulting differences were due to activation rather than deactivation.All results reported here were obtained by specifying a threshold ofPFWE < 0.05, and minimal cluster extent of 10 voxels. Data inspec-tion before the second-level analysis revealed that one of the partici-pants exhibited right hemisphere dominance, and she was thereforeexcluded from further analyses.

Lateralization of brain activation

The semantic categorization task presented to participants in the cur-rent study has been found to yield highly lateralized activation whenthe task was presented in participants’ L1 (Seghier et al., 2004).Performing semantic judgment tasks in L2 has been found to recruitregions in the right lateral frontal cortex, such as the precentralgyrus (R€uschemeyer et al., 2006) and middle frontal gyrus (MFG;Wartenburger et al., 2003), when compared with L1. This could be

expected to result in reduced lateralization for L2. In order to deter-mine whether lateralization of activation varied between languagesin the current study, LIs were calculated for each participant in eachlanguage, as described by Seghier et al. (2004). The following for-mula was used to calculate LIs:

LI ¼ ðvoxelsleft � voxelsrightÞ=ðvoxelsleft þ voxelsrightÞ

where voxelsleft and voxelsright are the number of voxels in a givenmap exceeding a selected significance threshold, residing in the leftand right hemispheres, respectively. The threshold used wasPunc. < 0.005, as in Seghier et al. (2004). LIs thus calculated wereentered into a repeated-measures ANOVA to examine effects oflanguage.

Regions of interest (ROIs) analyses

ROIs were defined in order to further examine effects of language.For this purpose, areas active during semantic categorization wereidentified based on the comparison between semantic categorizationand control across languages. Such a comparison yields regionsactive during categorization in any (or all) of the languages, andtherefore should not be expected to bias (as per the concerns raisedby Kriegeskorte et al., 2009) comparisons among languages, whichwere the objective of these analyses. The volume of the ROIs wasthresholded at 500 mm3. Due to extensive activation in the left fron-tal cortex, the activated voxels in this region were divided into sixregions, based on the automated anatomical labeling atlas (AAL;Tzourio-Mazoyer et al., 2002). In total, 11 ROIs were examined: (i)left triangularis; (ii) left opercularis; (iii) left MFG; (iv) left insula;(v) left precentral gyrus; (vi) left postcentral gyrus. Additional ROIswere defined in the: (vii) left middle temporal gyrus (MTG); (viii)left parietal (on the border between the parietal and occipital lobes);(ix) supplementary motor area (SMA); (x) right insula; and (xi) leftfusiform gyrus.Regarding the left fusiform ROI, this was defined based on exam-

ination of images of the differences between categorization andcontrol blocks generated for each of the participants separately.A threshold of Punc. < 0.001 was applied, with a cluster extent of 75voxels. This was done in order to reduce the probability that voxelsthat were strongly activated only in a minority of participants wouldbe included in the ROI, thereby ensuring the correct identification ofregion(s) selectively involved in the processing of written wordsamong a majority of participants. The resulting images were thenmasked so as to include only active voxels residing either in the leftinferior temporal gyrus, the left fusiform gyrus or the left inferioroccipital gyrus according to the AAL. The average of the maskedand thresholded images was calculated, and a threshold value of 4.87(corresponding to PFWE � 0.05) was applied to this image.Peri-stimulus time histograms were obtained for each participant,

in each region, in each language, using the MarsBaR toolbox forSPM (v0.43; Brett et al., 2002). The measure of brain activityselected was the mean signal change over the interval between 6and 30 s from the onset of blocks. These measures were used toexamine effects of language in the ROIs, as well as correlationsbetween activations in regions exhibiting effects of language.

Behavioral analysis

The individual mean RTs were computed from trials in which cor-rect responses were recorded, and were analysed in a two-way


Neural basis of diglossia in Arabic 3389

repeated-measures ANOVA with language and condition (semantic cat-egorization vs. control) as within-subject factors. A similar analysiswas performed on the individual accuracy rates computed separatelyfor each condition and language. All analyses were performed usingIBM SPSS Statistics software (v.19.0). In all ANOVAs, the Green-house–Geisser correction was applied for sphericity values lowerthan 0.75, and the Huynh–Feldt correction was applied for sphericityvalues greater than 0.75 (Field, 2005).

Results

Behavioral measures

Accuracy

The analysis of accuracy was computed as the percentage of correctresponses relative to the number of trials where a response wasobtained (Table 1). The two-way repeated-measures ANOVA per-formed on the individual values of accuracy showed that perfor-mance was higher in the control than in the experimental (semantic

categorization) condition (F1,24 = 59.3, P < 0.00001). There wasalso a significant main effect of language (F2,48 = 20.1,P < 0.00001) due to the fact that accuracy was higher in LA than inSA (Fisher’s LSD post hoc tests, P < 0.00001) and in Hebrew(P < 0.00001). The two-way interaction between language and taskwas also significant (F2,48 = 24.7, P < 0.00001). This interactionwas due to the fact that the language effect was significant only inthe activation (F2,48 = 26.3, P < 0.00001), but not in control blocks(P = 0.83; Table 1).

RTs

Mean RTs (�SD), based on trials with correct responses in eachcondition and language are presented in Table 1 (right column).The two-way repeated-measures ANOVA performed on the individualRTs using condition and language as within-subject factors showeda highly significant main effect of condition due to shorterresponses in control (mean = 663 ms) than in activation blocks(mean = 1032 ms, F1,24 = 266.9, P < 0.00001), and a main effectof language (F2,48 = 6.1, P < 0.005) due to faster responses in LAthan in SA (P < 0.005) and in Hebrew (P < 0.004), which did notdiffer from each other. The interaction between the two factorswas also significant (F2,48 = 9.7, P < 0.0003), due to the fact thatthe condition effect (i.e. difference between activation and control)was slightly smaller in LA than in the two other languages(Table 1).

fMRI analysis

Whole-brain analysis

The activation maps for the comparison between activation and con-trol conditions, analysed in all languages together, are displayed inFig. 1, which shows a dominant left hemisphere activation pattern.

Table 1. Accuracy and RTs on semantic categorization and control tasks,by language

Mean accuracyin % (�SD)

Mean RT inms (�SD)

LA activation 90.5 (7.8) 954 (131)SA activation 79.6 (10.8) 1074 (140)Hebrew activation 79.3 (7.9) 1068 (219)LA control 93.5 (4.1) 667 (116)SA control 93.6 (5.0) 656 (156)Hebrew control 93.9 (4.5) 666 (145)

LA, literary Arabic; RT, reaction time; SA, spoken Arabic.

Fig. 1. Regions more activated during semantic categorization compared with control, across languages.



As detailed in Table 2, these activations included, antero-posteriorlyin the left hemisphere, the inferior and middle frontal gyri (IFG andMFG, respectively), insula, precentral and postcentral gyri, the mid-dle and inferior temporal gyrus, and the fusiform and inferior occipi-tal gyri. In the right hemisphere, the activation included principallyparts of the right SMA, superior frontal gyrus, insula, IFG and cere-bellum. Figure 2A presents, based on activation vs. control in eachlanguage, a conjunction that shows the regions that were commonlyactivated in all languages. As displayed here and detailed inTable 3, the common activation was found almost exclusively in leftfrontal areas, including IFG, insula, precentral gyrus and MFG.Six possible pairwise comparisons between languages were con-

ducted [i.e. (i) LA vs. SA; (ii) LA vs. Hebrew; (iii) SA vs. LA; (iv)SA vs. Hebrew; (v) Hebrew vs. LA; (vi) Hebrew vs. SA]. Of these,only the comparison SA vs. LA yielded significant differences (atPFWE < 0.05) with stronger activation for SA than for LA (Fig. 2B;Table 4), though a threshold of Punc. < 0.001 did show differencesbetween Hebrew and LA in the left precentral gyrus and medial fron-tal and left occipital cortices, and between SA and Hebrew in the leftMTG and superior parietal lobule, and in the right inferior frontalcortex (precentral gyrus and IFG pars opercularis). Figure 2C pre-sents a superposition of the commonly activated voxels in the con-junction map (red areas, as in A), and those exhibiting significantdifferences between SA and LA (superposed in blue, as in B). Theregions that were more highly activated in SA relative to LAincluded the left IFG, precentral and postcentral gyri, and the leftinferior temporal gyrus (Table 4).

LIs

The effect of language on LIs was significant (F2,46 = 3.36,P < 0.05). LI for LA was greater than for SA. LIs for Hebrew werealso greater than for SA, though the difference was marginally sig-

nificant. Examination of the number of supra-threshold voxels ineach hemisphere yielded significant effects of language in bothhemispheres (F2,46 = 12.70, P < 0.001 and F2,46 = 10.39, P < 0.01for left and right hemispheres, respectively). In both hemispheresactivation was most extensive in SA, followed by Hebrew andfinally LA. However, in the left hemisphere differences between SAand LA were only marginally significant. Thus, it appears thatwhereas activation for Hebrew was proportionately more extensivein both hemispheres compared with LA, yielding similar values ofLI, activation in SA was particularly extensive in the right hemi-sphere, resulting in lower values.It is worthy of noting that LIs were generally lower than those

found by Seghier et al. (2004, 2008), with values (mean � SD) of0.59 � 0.22 for LA, 0.52 � 0.21 for SA and 0.59 � 0.2 forHebrew. However, setting the threshold at PFWE < 0.05 resulted inLIs similar to those reported by Seghier et al. (2004, 2008), with0.79 � 0.21 for LA, 0.74 � 0.2 for SA and 0.81 � 0.19 forHebrew. This may indicate that differences between LIs obtained inthe two studies may be due to the fact that in the current study par-ticipants were scanned using a 3T scanner, whereas Seghier et al.(2004) used a 1.5T scanner, which may have resulted in more vox-els, particularly in the right hemisphere, exceeding the threshold inthe current study. On the other hand, lower LIs may have to do withthe languages being examined; Al-Hamouri et al. (2005) reportstronger activation in the right hemisphere for Arabic compared withSpanish, and attribute this difference to ambiguities in decodingwritten words in Arabic presented without vowel diacritics. Thisinterpretation can be extended to unvowelled words in Hebrew, andmay be a contributing factor to the lower LIs found acrosslanguages in the current study.

Table 3. Conjunction of regions activated in all languages, when comparingsemantic categorization and control

Conjunction of categorization: LA, SA and Hebrew

Anatomical location (AAL) BA X; Y; Z Z K

Left IFG/pars triangularis 9 �42; 14; 26 Inf 169346 �52; 34; 12 Inf46 �42; 26; 22 Inf45 �50; 28; 16 Inf13 �42; 30; 6 5.66

Left insula 13 �34; 26; 4 5.38Left SMA 6 �4; 16; 58 6.18 169Right cerebellum 12; �78; �38 5.29 17

AAL, automated anatomical labeling; BA, Brodmann area; IFG, inferiorfrontal gyrus; LA, literary Arabic; SA, spoken Arabic; SMA, supplementarymotor area.

Table 2. Regions more active during semantic categorization comparedwith control, across languages

Semantic categorization vs. control

Anatomical location (AAL) BACoordinatesX; Y; Z Z-value

Clustersize

Left IFG pars triangularis 46 �44; 26; 22 12.22 368946 �52; 32; 12 12.1046 �44; 28; 14 11.509 �42; 14; 28 15.35

Left IFG pars opercularis 44 �52; 14;6 7.65Left precentral gyrus 6 �44; 4; 30 12.61

4 �52; �4; 48 8.27Left insula 13 �32; 24; 4 8.80SMA 6 �4; 16; 56 10.14 776

6 10; 18; 50 5.45Left inferior temporal & occipitalgyri/fusiform gyrus

37 �44; �56; �12 6.68 55�44; �66; �14 5.45

Left MTG 22 �52; �42; 4 6.08 151�60; �36; 4 5.66

Left inferior parietal/middleoccipital gyrus

7 �28; �66; 40 7.08 119

Left inferior & middle occipitalgyri

18 �28; �96; �12 5.99 47�34; �92; �4 5.78

Right insula/IFG (parstriangularis)

32; 26; 4 7.79 187

Right cerebellum 10; �78; �38 8.31 196

AAL, automated anatomical labeling; BA, Brodmann area; IFG, inferiorfrontal gyrus; MTG, middle temporal gyrus; SMA, supplementary motorarea.

Table 4. Regions more active in SA compared with LA during semanticcategorization

Semantic categorization: SA vs. LA

Anatomical location (AAL) BA X; Y; Z Z K

IFG (pars opercularis) 9 �54; 8; 22 5.92 370Precentral gyrus/postcentral gyrus 6 �50; 0; 38 6.19Inferior temporal gyrus 37 �50; �64; �8 5.21 15

AAL, automated anatomical labeling; BA, Brodmann area; IFG, inferiorfrontal gyrus; LA, literary Arabic; SA, spoken Arabic.



A

B C

Fig. 2. (A) Regions activated in all three languages (red). (B) Regions activated more strongly in spoken Arabic (SA) compared with literary Arabic (LA;blue). (C) Superposition of (A) and (B).

Fig. 3. Regions of interest (ROIs) exhibiting significant effects of language. Error bars indicate limits of 95% confidence intervals about the mean.



ROI analyses

Significant effects of language were found in four regions: left opercu-laris (F2,46 = 4.7, P < 0.05); left precentral (F2,46 = 4.16, P < 0.05);left parietal (F2,46 = 4.46, P < 0.05); and left fusiform (F2,46 = 3.51,P < 0.05). Results of pairwise comparisons indicated that activationfor SA was stronger than for LA in all four regions, whereas activationfor Hebrew did not differ significantly from activation for LA. Fur-thermore, in two of the four regions (left precentral and left parietal),activation for SA was also stronger than activation for Hebrew, andthe difference between SA and Hebrew was marginally significantin the left fusiform gyrus (Fig. 3). Marginally significant effects oflanguage were also found in the left triangularis (F2,46 = 2.67,P = 0.081, SA > LA) and postcentral gyrus (F2,46 = 3.02, P = 0.061,SA > LA). Additional analyses of the regions, which best differenti-ated SA from LA and Hebrew, showed that the activity duringprocessing of SA yielded significant correlations between the precen-tral gyrus and the left parietal, and between the precentral and the fusi-form gyrus (r = 0.68, P < 0.01; r = 0.63, P < 0.01, respectively,after correction for multiple comparisons). It should be noted that thedifferences found here between SA and LA in the parietal lobe hadnot been evident in the whole-brain analysis. However, when in thisanalysis the threshold was lowered to Punc. < 0.001 (instead ofPFWE < 0.05), a significant cluster did indeed emerge.

Discussion

In the current study, our objective was to compare the processing ofvisually presented SA and LA words in Arabic–Hebrew bilinguals.For this purpose, participants performed a semantic categorizationtask on written words in LA, SA and Hebrew. We predicted that theprocessing of SA and LA words would reflect their history of acquisi-tion and patterns of use. Thus, we expected that SA, while being thefirst language acquired by Arabic speakers but usually not encoun-tered in the written form, would show response and activation pat-terns mimicking either L2 words or low-frequency/unfamiliar words.In contrast, because LA is the first written form acquired and morefrequently used in writing, we predicted superior behavioral perfor-mance for LA words relying on the classical left language network.Our results showed that accuracy was high for all languages, con-

firming the ability of the participants to correctly identify the stim-uli. As predicted, categorization of LA words was faster and moreaccurate than categorization of words in either SA or Hebrew, whichdid not differ from each other.In terms of brain activity, whole-brain analysis first showed that

categorization vs. control conditions across all three language varie-ties revealed a classical pattern of activation consisting mainly ofleft hemisphere areas (Seghier et al., 2004, 2008). Conjunctionanalysis between the three language varieties showed a stronginvolvement in the left hemisphere of the IFG, MFG, insula andSMA. When contrasting the three varieties, differences wereobserved only between SA vs. LA, with stronger activation for SAin left frontal and temporal areas.We followed this whole-brain analysis with an ROI analysis, in

which ROIs were selected based on areas most strongly distinguish-ing between categorization and control across languages. In thisanalysis, a more complex pattern emerged. SA generated strongeractivation than LA in four of the regions examined in the left hemi-sphere, namely opercularis, precentral, parietal and fusiform. Inaddition, SA generated stronger activation than Hebrew in left pre-central and left parietal areas. In all of these regions, activation forLA did not differ from activation for Hebrew.

Lateralization of activation was examined using LIs. In theseanalyses, lateralization for LA was stronger than for SA, and lateral-ization for Hebrew was similar to that found for LA, though onlymarginally stronger than for SA. This pattern was due to differencesin the extent of activated voxels in each hemisphere in each of thelanguages. Activation in both hemispheres, but particularly in theright one, was most extensive for SA. Activation for Hebrew wasless extensive than for SA, but more extensive than for LA. Theseresults were supported by the fact that when a more lenient thresh-old was applied to whole-brain analyses, differences betweenHebrew and LA, and between SA and Hebrew emerged, includingdifferences in the right inferior frontal cortex.Results of the current study reflect participants’ higher proficiency

in LA compared with Hebrew. This is particularly evident in partici-pants’ performance but, also, albeit more subtly, in measures ofbrain activation. Thus, comparisons between Hebrew and LA usinga threshold of Punc. < 0.001 revealed differences in the left precen-tral gyrus and in the SMA, while counting active voxels in eachhemisphere reveals more extensive activation bilaterally for Hebrew.Stronger activation in the bilateral precentral gyri and SMA has

been reported during reading of sentences in L2 compared with L1,in the context of semantic and grammatical acceptability judgmenttasks (R€uschemeyer et al., 2006). Stronger activation in the SMAhas also been found during reading of words in a less proficient L1compared with a more proficient L2 (Meschyan & Hernandez,2006). Finally, Wartenburger et al. (2003) report proficiency-relateddifferences in activation in the left IFG during a semantic judgmenttask, at a location adjacent to the foci of differences in the precentralgyrus in the comparison between Hebrew and LA.Regarding the observed differences between SA and LA in terms

of performance and brain activity, these may be related to the rela-tive unfamiliarity of written words in SA. Given the fact that wordsin SA are less often encountered in the written form than words inLA, effects of familiarity for SA word forms are to be expected invisual presentation. Such differences would most likely emerge inthe left fusiform gyrus, where word frequency has been found toaffect activation during reading (Joubert et al., 2004; Kronbichleret al., 2004). In this context, differences in lateralization of activa-tion between SA and LA, which are related to more extensive acti-vation in the right inferior frontal cortex for SA, may indicate thatparticipants had to effortfully avoid interpreting words in SA aswords in LA. Previous studies reported in the involvement of theright IFG during increased processing demands and inhibition dur-ing ‘go/no-go’ tasks (Chikazoe et al., 2007; Lenartowicz et al.,2011). Additionally, presentation of written words in SA may haveresulted in enhanced sub-lexical phonological processing. Evidenceof such processes during reading of written words in SA has beenpreviously presented by Bentin & Ibrahim (1996).Results concerning pseudoword reading are of relevance to the

investigation of sub-lexical phonological processing, as reading ofpseudowords is thought to require reliance on such processing.Greater activation for pseudowords compared with words has beenreported in the left IFG during overt (Carreiras et al., 2007; Heimet al., 2013) and silent (Joubert et al., 2004) reading. Activation inthe left parietal cortex [intra-parietal sulcus (IPS)] has been reportedduring both silent and overt reading (Dietz et al., 2005) as well.Additionally, activation in left parietal and inferior frontal regionshas been found to be correlated with reading proficiency (Jobardet al., 2011). The authors attributed the latter effect to phonologicalprocessing. Finally, activity in the left parietal lobe has been foundto be associated with articulation (e.g. stimulation of the IPS canevoke such intentions; Desmurget et al., 2009), supporting the



possibility that activation in the left IPS reflects stronger reliance onphonological processing for written words in SA.Alternatively, in light of the findings of Chou and colleagues

(Chou et al., 2006a, 2009), stronger activation in the left IPS mayreflect stronger semantic associations among words in SA. In thesestudies, activation in the left IPS was found to increase with seman-tic relatedness (Chou et al., 2006a, 2009). Additionally, when thetasks were presented to children between 9 and 15 years old (Chouet al., 2006b), activation in the left IPS was found to increase withage. This interpretation therefore attributes differences in behavioralmeasures between LA and SA, as well as differences in activationin the left fusiform, precentral and IFG, to the relative unfamiliarityof written word forms in SA, as did the one presented above. Oncewords have been successfully decoded, however, increased activa-tion in the left IPS is taken to reflect stronger connections to seman-tic representations. Such an interpretation would be consistent withwords in SA being early acquired and highly familiar, despite thefact that visual presentation of these words is unusual.The analyses presented demonstrate, therefore, that specific

regions, namely the left opercularis and the left fusiform, were morestrongly activated in SA when compared with LA, but did not dis-tinguish the activation of SA and Hebrew. This finding suggests thatfor the unique population of native Arabic speakers, who are bothdiglossic and bilingual, both the first acquired SA and the lateracquired Hebrew at times ‘look’ like an L2 in the written modality.Our findings contrast with previous research using auditory stim-

uli, which indicated that SA words and LA words are processed asL1 and L2, respectively (Ibrahim & Aharon-Peretz, 2005; Ibrahim,2009). This apparent contradiction stems from the uniqueness of thediglossic situation where language status (as a first or second lan-guage) is tightly linked to the modality of presentation. Thus, SA isthe first acquired variety, used mainly in spoken language, andtherefore occupies a privileged position in processing auditory stim-uli. In contrast, LA is acquired later in life, but is then used almostexclusively in the written modality, leading to an advantage invisual word processing, as demonstrated in the current study.Finally, the question remains as to whether the current findingsmight be better explained in terms of bilingualism and language sta-tus, or in terms of effects of familiarity of word forms of the samelanguage across modalities of presentation. Further functional inves-tigations are needed to assess, for example, how low-frequency LAwords compare with SA words in the written modality, and howdominance in the auditory vs. visual modality modulates brain acti-vation patterns during SA and LA processing.

Acknowledgements

The authors thank all the participants for their participation in this study.This research was supported by the Israeli Science Foundation (Grant no.623/11 awarded to A.K. and A.P.) and by the Edmond J. Safra BrainResearch Center for the Study of Learning Disabilities. The authors declarethat there is no conflict of interest related to the work presented in this paper.This article is dedicated to the memory of Prof. Zvia Breznitz, head of theEdmond J. Safra Brain Research Center for the Study of Learning Disabil-ities, who passed away on May 18th, 2014.

Abbreviations

AAL, automated anatomical labeling; fMRI, functional magnetic resonanceimaging; IFG, inferior frontal gyrus; IPS, intra-parietal sulcus; LA, literaryArabic; LI, laterality index; MFG, middle frontal gyrus; MTG, middle tem-poral gyrus; ROI, region of interest; RT, reaction time; SA, spoken Arabic;SMA, supplementary motor area; SR, semantically related; SU, semanticallyunrelated; TE, echo time; TR, repetition time.

References

Abu-Rabia, S. (2000) Effects of exposure to literary Arabic on reading com-prehension in a diglossic situation. Read. Writ., 13, 147–157.

Abu-Rabia, S. (2001) The role of vowels in reading Semitic scripts: datafrom Arabic and Hebrew. Read. Writ., 14, 39–59.

Al-Hamouri, F., Maest�u, F., del R�ıo, D., Fern�andez, S., Campo, P., Almude-na, C., Garc�ıa, E., Gonz�ales-Marqu�es, J. & Ortiz, T. (2005) Brain dynam-ics of Arabic reading: a magnetoencephalography study. NeuroReport, 16,1861–1864.

Ayari, S. (1996) Diglossia and illiteracy in the Arab world. Cult. Curriculum,9, 243–253.

Bentin, S. & Ibrahim, R. (1996) New evidence for phonological processingduring visual word recognition: the case of Arabic. J. Exp. Psychol.Learn., 22, 309–323.

Boudelaa, S. & Marslen-Wilson, W.D. (2013) Morphological structure in theArabic mental lexicon: parallels between standard and dialectal Arabic.Lang. Cognitive Proc., 28, 1453–1473.

Brett, M., Anton, J.L., Valabregue, R. & Polin, J.B. (2002) Region of inter-est analysis using an SPM toolbox. Presented at the 8th International Con-ference on Functional Mapping of the Human Brain, Sendai, Japan.Available on CD-ROM in Neuroimage, 16.

Carreiras, M., Mechelli, A., Est�evez, A. & Price, C.J. (2007) Brain activationfor lexical decision and reading aloud: two sides of the same coin? J. Cog-nitive Neurosci., 19, 433–444.

Chikazoe, J., Konishi, S., Asari, T., Jimura, K. & Miyashita, Y. (2007) Acti-vation of right inferior frontal gyrus during response inhibition acrossresponse modalities. J. Cognitive Neurosci., 19, 69–80.

Chou, T.L., Booth, J.R., Burman, D.D., Bitan, T., Bigio, J.D., Lu, D. &Cone, N.E. (2006a) Developmental changes in the neural correlates ofsemantic processing. NeuroImage, 29, 1141–1149.

Chou, T.L., Booth, J.R., Bitan, T., Burman, D.D., Bigio, J.D., Cone, N.E.,Lu, D. & Cao, F. (2006b) Developmental and skill effects on the neuralcorrelates of semantic processing to visually presented words. Hum. BrainMapp., 27, 915–924.

Chou, T.L., Chen, C.W., Wu, M.Y. & Booth, J.R. (2009) The role of theinferior frontal gyrus and inferior parietal lobule in semantic processing ofChinese characters. Exp. Brain Res., 198, 465–475.

Desmurget, M., Reilly, K.T., Richard, N., Szathmari, A., Mottolese, C. &Sirigu, A. (2009) Movement intention after parietal cortex stimulation inhumans. Science, 324, 811–813.

Dietz, N.A., Jones, K.M., Gareau, L., Zeffiro, T.A. & Eden, G.F. (2005)Phonological decoding involves left posterior fusiform gyrus. Hum. BrainMapp., 26, 81–93.

Eviatar, Z. & Ibrahim, R. (2000) Bilingual is as bilingual does: meta-linguisticabilities of Arabic-speaking children. Appl. Psycholinguist., 21, 451–471.

Ferguson, C. (1959) Diglossia. Word, 15, 325–340.Field, A. (2005) Statistics Using SPSS, 2nd Edn. SAGE Publications Ltd,London.

Friston, K.J., Holmes, A.P., Poline, J.B., Grasby, P.J., Williams, S.C., Frac-kowiak, R.S. & Turner, R. (1995) Analysis of fMRI time-series revisited.NeuroImage, 2, 45–53.

Friston, K., Ashburner, J., Kiebel, S., Nichols, T. & Penny, W. (2007) Statis-tical Parametric Mapping: The Analysis of Functional Brain Images.Academic Press, London.

Gollan, T.H., Forster, K.I. & Frost, R. (1997) Translation priming with dif-ferent scripts: masked priming with cognates and noncognates in Hebrew-English Bilinguals. J. Exp. Psychol. Learn., 23, 1122–1139.

Heim, S., Wehnelt, A., Grande, M., Huber, W. & Amunts, K. (2013) Effectsof lexicality and word frequency on brain activation in dyslexic readers.Brain Lang., 125, 194–202.

Ibrahim, R. (2009) The cognitive basis of diglossia in Arabic: evidence froma repetition priming study within and between languages. Psychol. Res.Behav. Manage., 2, 93–105.

Ibrahim, R. & Aharon-Peretz, J. (2005) Is Literary Arabic a second languagefor native Arab speakers? Evidence from semantic priming study. J. Psy-cholinguist. Res., 34, 51–70.

Jobard, G., Vigneau, M., Simon, G. & Tzourio-Mazoyer, N. (2011) Theweight of skill: inter-individual variability of reading related brain activa-tion patterns in fluent readers. J. Neurolinguist., 24, 113–132.

Joubert, S., Beauregard, M., Walter, N., Bourgouin, P., Beaudoin, P.,Leroux, J.M., Karama, S. & Roch Lecours, A. (2004) Neural correlates oflexical and sublexical processes in reading. Brain Lang., 89, 9–20.

Keatley, C.W., Spinks, J.A. & de Gelder, B. (1994) Asymmetrical cross-language priming effects. Mem. Cognition, 22, 70–84.



Khateb, A., Michel, C.M., Pegna, A.J., O’Dochartaigh, S.D., Landis, T. &Annoni, J.M. (2003) Processing of semantic categorical and asso-ciative relations: an ERP mapping study. Int. J. Psychophysiol., 49, 41–55.

Kriegeskorte, N., Simmons, W.K., Bellgowan, P.S.F. & Baker, C.I. (2009)Circular analysis in systems neuroscience: the dangers of double dipping.Nat. Neurosci., 12, 535–540.

Kronbichler, M., Hutzler, F., Wimmer, H., Mair, A., Staffen, W. & Ladurner,G. (2004) The visual word form area and the frequency with which wordsare encountered: evidence from a parametric fMRI study. NeuroImage, 21,946–953.

Lenartowicz, A., Verbruggen, F., Logan, G.D. & Poldrack, R.A. (2011) Inhi-bition-related activation in the right inferior frontal gyrus in the absence ofinhibitory cues. J. Cognitive Neurosci., 23, 3388–3399.

Levin, I., Saiegh-Haddad, E., Hende, N. & Ziv, M. (2008) Early literacy inArabic: an intervention among Israeli Palestinian kindergartners. Appl. Psy-cholinguist., 29, 413–436.

Meschyan, G. & Hernandez, A.E. (2006) Impact of language proficiency andorthographic transparency on bilingual word reading: an fMRI investiga-tion. NeuroImage, 29, 1135–1140.

R€uschemeyer, S.A., Zysset, S. & Friederici, A.D. (2006) Native and non-nativereading of sentences: an fMRI experiment. NeuroImage, 31, 354–365.

Saiegh-Haddad, E. (2003) Linguistic distance and initial reading acquisition:the case of Arabic diglossia. Appl. Psycholinguist., 24, 431–451.

Saiegh-Haddad, E. (2005) Correlates of reading fluency in Arabic: diglossicand orthographic factors. Read. Writ., 18, 559–582.

Saiegh-Haddad, E., Levin, I., Hende, N. & Ziv, M. (2011) The linguisticaffiliation constraint and phoneme recognition in diglossic Arabic. J. ChildLang., 38, 297–315.

Seghier, M.L., Lazeyras, F., Pegna, A.J., Annoni, J.M., Zimine, I., Mayer,E., Michel, C.M. & Khateb, A. (2004) Variability of fMRI activation dur-ing a phonological and semantic language task in healthy subjects. Hum.Brain Mapp., 23, 140–155.

Seghier, M.L., Lazeyras, F., Pegna, A.J., Annoni, J.M. & Khateb, A. (2008)Group analysis and the subject factor in functional magnetic resonanceimaging: analysis of fifty right-handed healthy subjects in a semantic lan-guage task. Hum. Brain Mapp., 29, 461–477.

Tzourio-Mazoyer, N., Landeau, B., Papathanassiou, D., Crivello, F., Etard,O., Delcroix, N., Mazoyer, B. & Joliot, M. (2002) Automated anatomi-cal labeling of activations in SPM using a macroscopic anatomicalparcellation of the MNI MRI single-subject brain. NeuroImage, 15,273–289.

Wartenburger, I., Heekeren, H.R., Abutalebi, J., Cappa, S.F., Villringer, A.& Perani, D. (2003) Early setting of grammatical processing in the bilin-gual brain. Neuron, 37, 159–170.

Worsley, K.J. & Friston, K.J. (1995) Analysis of fMRI time-series revisited–again. NeuroImage, 2, 173–181.



Documents

When first language is not first: an functional magnetic resonance … › articles › prior › When first... · 2018-12-03 · COGNITIVE NEUROSCIENCE When ﬁrst language is not